为了验证雷达目标辨识网络存在风险,提升基于辨识网络的雷达目标辨识效果,文中研究了不同雷达目标辨识网络的对抗性样本。针对雷达目标辨识网络中的卷积神经网络和分解卷积神经网络,建立对抗性样本生成模型,按照该模型生成对抗性样本,并对生成的结果进行分析总结。实测数据处理结果表明,雷达目标辨识网络存在潜在风险。

For verifying the risky of radar target discrimination network and improving the effect of radar target discrimination based on discrimination network, we do research on adversarial examples of different radar target discrimination network. Focusing on two radar target discrimination network convolutional neural network and factorized convolutional neural network, we construct the model of adversarial example and generate adversarial example using the model, and do analysis of the result. The result of the experiment shows that the two discrimination networks have risk.

TN974